Abstract: In order to fully utilize the regulation capability of multiple types of regulation resources in the distribution network and improve the zonal autonomy capability of the distribution network under the high proportion of new energy access, this paper proposes a multi-zonal cooperative control strategy for distribution network voltage based on multi-agent deep reinforcement learning (MADRL). The problem of partition cooperative voltage control in distribution network is modeled using a multi-agent approach. Subsequently, an improved counterfactual multi-agent soft actor-critic (COMASAC) deep reinforcement learning model is applied to solve the zonal cooperative voltage control problem in distribution networks.Finally, simulation examples using typical day operational data from actual distribution networks demonstrate the advantages of the proposed multi-agent deep reinforcement learning method in improving voltage stability and reducing network losses in distribution networks.

    Key words: multi-agent; deep reinforcement learning; distribution network voltage; partition cooperative control; network loss